Measurement and prediction of small molecule retention by Gram-negative bacteria based on a large-scale LC/MS screen.

The challenge of assessing intracellular accumulation represents a major hurdle to the discovery of new antibiotics with Gram-negative activity. To address this, a high-throughput assay was developed to measure compound uptake and retention in Escherichia coli using LC/MS. 13,056 diverse small molecules were screened with two isogenic E. coli strains, a wild-type and a TolC-deleted mutant. Cell-associated concentrations of 8,410 compounds were determined and 6,416 compounds were classified either as retention-positive or -negative, with 45% (2,885) positives in the TolC mutant. Of these, 60% were not retained in the wild-type strain, indicating efficient efflux. No individual structural feature or physicochemical property explained the retention behavior. Machine learning (ML) models were trained using these results, and a gradient-boosted-tree model using 36 physicochemical compound descriptors proved most accurate. The ML model demonstrated robust performance across similar and dissimilar molecule subsets, showcasing its strong generalization and real-world predictive potential. An experimental validation of the tool was performed with a set of 540 new compounds and correctly predicted retention-positive cases in 77.8% and retention-negative in 74.4%. This assay and prediction tool could enhance Gram-negative antibiotic discovery, aiding in screening library design, computational structure-based drug design, and exploration of chemical space before synthesis.